|
|
Results in Probability Scale |
|
Message/Author |
|
|
Hello: Following is a simple model I estimated using both ML and Bayes (50000 biterations, PSR ~ 1.00 for last 45000). The estimates of the thresholds are quite different for the two estimation procedures and, as a consequence, so are the probabilities estimated from the threshold estimates. However with ML the estimates in the Results in Probability Scale (RPS) are equal to those computed from the threshold estimates. With Bayes this is not true. Further, the RPS for ML and Bayes almost identical (one estimate differs in the 3rd place) and for RPS ML standard errors and the Bayes posterior SD are identical. Are the RPS results for Bayes not computed from the threshold estimates or is the code incorrect for using BAYES? Thanks, Jamie CODE: Model: [DV$1] (THRESH1); [DV$2] (THRESH2); model constraint: new P_L P_M P_H ; P_H = 1/(1 + EXP(Thresh2)); P_M= 1/(1 + EXP(Thresh1)) - 1/(1 + EXP(Thresh2)); P_L = 1- 1/(1 + EXP(Thresh1)); |
|
|
ML is logistic. Bayes is probit. The way to compute probabilities differs. See Chapter 14 of the user's guide. |
|
Back to top |
|
|